Makefile: Add a pcheck option to run tests in parallel

Running tests in parallel is much faster, e.g. 15 seconds to run the tests
on sandbox (only), instead of 100 seconds (on a 16-core machine). Add a
'make pcheck' option to access this feature.

Note that the tools/ tests still run each tool's tests once after the
other, although within that, they do run in parallel. So for example,
the buildman tests run in parallel, then the binman tests run in
parallel. There would be a signiificant advantage to running them all
in parallel together, but that would require a large amount of
refactoring, e.g. with more use of pytest fixtures.

Update the documentation to represent the current state.

Signed-off-by: Simon Glass <sjg@chromium.org>

diff --git a/Makefile b/Makefile
index 27375f033f434a497793e271512507678b7e6e5d..0b8079cf81def841d3d5d35b6a38eb8c67179114 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -521,8 +521,8 @@ env_h := include/generated/environment.h
 
 no-dot-config-targets := clean clobber mrproper distclean \
                         help %docs check% coccicheck \
-                        ubootversion backup tests check qcheck tcheck pylint \
-                        pylint_err
+                        ubootversion backup tests check pcheck qcheck tcheck \
+                        pylint pylint_err
 
 config-targets := 0
 mixed-targets  := 0
@@ -2364,6 +2364,7 @@ help:
        @echo  'Test targets:'
        @echo  ''
        @echo  '  check           - Run all automated tests that use sandbox'
+       @echo  '  pcheck          - Run quick automated tests in parallel'
        @echo  '  qcheck          - Run quick automated tests that use sandbox'
        @echo  '  tcheck          - Run quick automated tests on tools'
        @echo  '  pylint          - Run pylint on all Python files'
@@ -2409,6 +2410,9 @@ help:
 tests check:
        $(srctree)/test/run
 
+pcheck:
+       $(srctree)/test/run parallel
+
 qcheck:
        $(srctree)/test/run quick
 

diff --git a/doc/develop/py_testing.rst b/doc/develop/py_testing.rst
index 06f919609b5785eeadf1d54c3ae5ae3b9b87f60a..92fbd22721942898705160b82f865ed38f1507a8 100644 (file)
--- a/doc/develop/py_testing.rst
+++ b/doc/develop/py_testing.rst
@@ -121,31 +121,36 @@ more options.
 Running tests in parallel
 ~~~~~~~~~~~~~~~~~~~~~~~~~
 
-Note: This does not fully work yet and is documented only so you can try to
-fix the problems.
+Note: Not all tests can run in parallel at present, so the usual approach is
+to just run those that can.
 
 First install support for parallel tests::
 
+    sudo apt install python3-pytest-xdist
+
+or:::
+
     pip3 install pytest-xdist
 
-Then build sandbox in a suitable build directory. It is not possible to use
-the --build flag with xdist.
+Then run the tests in parallel using the -n flag::
 
-Finally, run the tests in parallel using the -n flag::
+    test/py/test.py -B sandbox --build --build-dir /tmp/b/sandbox -q -k \
+        'not slow and not bootstd and not spi_flash' -n16
 
-    # build sandbox first, in a suitable build directory. It is not possible
-    # to use the --build flag with -n
-    test/py/test.py -B sandbox --build-dir /tmp/b/sandbox -q -k 'not slow' -n32
+You can also use `make pcheck` to run all tests in parallel. This uses a maximum
+of 16 threads, since the setup time is significant and there are under 1000
+tests.
 
-At least the following non-slow tests are known to fail:
+Note that the `test-log.html` output does not work correctly at present with
+parallel testing. All the threads write to it at once, so it is garbled.
 
-- test_fit_ecdsa
-- test_bind_unbind_with_uclass
-- ut_dm_spi_flash
-- test_gpt_rename_partition
-- test_gpt_swap_partitions
-- test_pinmux_status
-- test_sqfs_load
+Note that the `tools/` tests still run each tool's tests once after the other,
+although within that, they do run in parallel. So for example, the buildman
+tests run in parallel, then the binman tests run in parallel. There would be a
+significant advantage to running them all in parallel together, but that would
+require a large amount of refactoring, e.g. with more use of pytest fixtures.
+The code-coverage tests are omitted since they cannot run in parallel due to a
+Python limitation.
 
 
 Testing under a debugger

diff --git a/doc/develop/testing.rst b/doc/develop/testing.rst
index 1abe4d7f0f0e9c54aa179b238dc0e97edd7e796c..5afeb42f69682e2a9ce616cb6cc0cc98c842e931 100644 (file)
--- a/doc/develop/testing.rst
+++ b/doc/develop/testing.rst
@@ -28,8 +28,12 @@ run. Type this::
 
     make tcheck
 
+You can also run a selection tests in parallel with::
+
+    make pcheck
+
 All of the above use the test/run script with a paremeter to select which tests
-are run.
+are run. See :doc:`py_testing` for more information.
 
 
 Sandbox

diff --git a/test/run b/test/run
index 0c8aeb5b0afd4621643730bc219c55335df70b30..810b47e08d0c5acbb45e3e5387811576669f94c4 100755 (executable)
--- a/test/run
+++ b/test/run
@@ -14,27 +14,46 @@ run_test() {
 }
 
 # Select test attributes
+ut_mark_expr=test_ut
 if [ "$1" = "quick" ]; then
        mark_expr="not slow"
+       ut_mark_expr="test_ut and not slow"
        skip=--skip-net-tests
 fi
 
 [ "$1" == "tools" ] && tools_only=y
 
+if [ "$1" = "parallel" ]; then
+       if ! echo 'import xdist' | python3 2>/dev/null; then
+               echo "Please install python3-pytest-xdist - see doc/develop/py_testing.rst"
+               exit 1
+       fi
+       jobs="$(($(nproc) > 16 ? 16 : $(nproc)))"
+       para="-n${jobs} -q"
+       prompt="Building and..."
+       skip=--skip-net-tests
+       mark_expr="not slow and not bootstd and not spi_flash"
+       ut_mark_expr="test_ut and not slow and not bootstd and not spi_flash"
+       echo "Note: test log is garbled with parallel tests"
+fi
+
 failures=0
 
 if [ -z "$tools_only" ]; then
        # Run all tests that the standard sandbox build can support
-       run_test "sandbox" ./test/py/test.py --bd sandbox --build \
+       echo "${prompt}"
+       run_test "sandbox" ./test/py/test.py --bd sandbox --build ${para} \
                -k "${mark_expr}"
 fi
 
 # Run tests which require sandbox_spl
-run_test "sandbox_spl" ./test/py/test.py --bd sandbox_spl --build \
+echo "${prompt}"
+run_test "sandbox_spl" ./test/py/test.py --bd sandbox_spl --build ${para} \
                -k 'test_ofplatdata or test_handoff or test_spl'
 
 # Run the sane tests with sandbox_noinst (i.e. without OF_PLATDATA_INST)
-run_test "sandbox_spl" ./test/py/test.py --bd sandbox_noinst --build \
+echo "${prompt}"
+run_test "sandbox_spl" ./test/py/test.py --bd sandbox_noinst --build ${para} \
                -k 'test_ofplatdata or test_handoff or test_spl'
 
 if [ -z "$tools_only" ]; then
@@ -42,8 +61,9 @@ if [ -z "$tools_only" ]; then
        # build which does not enable CONFIG_OF_LIVE for the live device tree, so we can
        # check that functionality is the same. The standard sandbox build (above) uses
        # CONFIG_OF_LIVE.
+       echo "${prompt}"
        run_test "sandbox_flattree" ./test/py/test.py --bd sandbox_flattree \
-               --build -k test_ut
+               ${para} --build -k "${ut_mark_expr}"
 fi
 
 # Set up a path to dtc (device-tree compiler) and libfdt.py, a library it
@@ -64,10 +84,14 @@ run_test "dtoc" ./tools/dtoc/dtoc -t
 # This needs you to set up Python test coverage tools.
 # To enable Python test coverage on Debian-type distributions (e.g. Ubuntu):
 #   $ sudo apt-get install python-pytest python-coverage
-export PATH=$PATH:${TOOLS_DIR}
-run_test "binman code coverage" ./tools/binman/binman test -T
-run_test "dtoc code coverage" ./tools/dtoc/dtoc -T
-run_test "fdt code coverage" ./tools/dtoc/test_fdt -T
+
+# Code-coverage tests cannot run in parallel, so skip them in that case
+if [ -z "${para}" ]; then
+       export PATH=$PATH:${TOOLS_DIR}
+       run_test "binman code coverage" ./tools/binman/binman test -T
+       run_test "dtoc code coverage" ./tools/dtoc/dtoc -T
+       run_test "fdt code coverage" ./tools/dtoc/test_fdt -T
+fi
 
 if [ $failures == 0 ]; then
        echo "Tests passed!"